Feature request (Again)

[JuanTutors]

My request is that during P-1 testing for 100M mersennes, the GCD is done more frequently since P-1 testing takes SO LONG to complete for these numbers.

I made this feature request before for 10M mersennes, but it's more relevant now that we can do primality tests of 100M mersennes in an appreciable amount of time.

[cheesehead]

Originally Posted by dominicanpapi82

My request is that during P-1 testing for 100M mersennes, the GCD is done more frequently since P-1 testing takes SO LONG to complete for these numbers.

It could be done. The P-1 limit selection routine could be modified to calculate whether an intermediate GCD could save enough time, often enough, to reduce the estimated total P-1 time enough to offset the estimated extra GCD time.

However, it would save little time -- I guess something on the order of: 4% of such P-1 runs could save half their total time (or 3% of the runs could save 2/3 of their total time, and 1.5% of the remainder could save 1/3 of their time), for overall P-1 throughput improvement of only about 2-3%.

[JuanTutors]

In the case of 10M mersennes, I would have read that and said, OK, that 2%-3% is not really much or even worth it to the user. However, for a 100M mersenne, which takes so long to complete a full test, it's really worth checking GCDs intermittently for the user.

quote=cheesehead;145210]It could be done. The P-1 limit selection routine could be modified to calculate whether an intermediate GCD could save enough time, often enough, to reduce the estimated total P-1 time enough to offset the estimated extra GCD time.

However, it would save little time -- I guess something on the order of: 4% of such P-1 runs could save half their total time (or 3% of the runs could save 2/3 of their total time, and 1.5% of the remainder could save 1/3 of their time), for overall P-1 throughput improvement of only about 2-3%.[/quote]

[cheesehead]

Originally Posted by dominicanpapi82

In the case of 10M mersennes, I would have read that and said, OK, that 2%-3% is not really much or even worth it to the user. However, for a 100M mersenne, which takes so long to complete a full test, it's really worth checking GCDs intermittently for the user.

If you really want to do that to satisfy your impatience, you can always:

1) Stop the P-1 right after it starts, so you can see what limits Prime95 chose. (Prime95 will write a P-1 save file.)

2) Ahead of the "Test=" line, insert "Pminus1=" lines with appropriate numbers to cause it to do the GCD as often as you like, finishing up with a final "Pminus1=" with the Prime95-chosen limits.

3) On the "Test=" line, change the third parameter from 0 to 1. (Format is Test=exponent,how_far_factored,has_been_pminus1ed) If Prime95 did trial factoring before the P-1 step, make sure the second parameter matches the limit to which the trial factoring went, rather than some lower value where TF started.

4) Start Prime95 again. It will use the P-1 save file to continue from where it left off, toward the limits on your first "Pminus1=" line.

[cheesehead]

Originally Posted by cheesehead

2) Ahead of the "Test=" line, insert "Pminus1=" lines with appropriate numbers to cause it to do the GCD as often as you like, finishing up with a final "Pminus1=" with the Prime95-chosen limits.

... in the worktodo file, that is.

- - -

dominicanpapi82,

I am struck by the contrast between your "we can do primality tests of 100M mersennes in an appreciable amount of time" and your "P-1 testing takes SO LONG to complete for these numbers".

Is a 2-3% savings on a bunch of P-1 tests all that significant compared to the much longer total of L-L test times for that bunch of numbers?

Maybe I should start one to see what the time estimates are.

[JuanTutors]

I think you're trying to hard to parse words. Here's the thing. With a 10M mersenne, we're talking spending a few minutes to possibly save a few hours. With 100M mersennes, we're talking spending a few more minutes to possibly save days. That's a significant difference. If you don't think scale is relevant, then we're going to disagree.

[cheesehead]

Originally Posted by dominicanpapi82

I think you're trying to hard to parse words. Here's the thing. With a 10M mersenne, we're talking spending a few minutes to possibly save a few hours. With 100M mersennes, we're talking spending a few more minutes to possibly save days. That's a significant difference. If you don't think scale is relevant, then we're going to disagree.

Can you quantify "few", "more", "significant", and so forth?

[JuanTutors]

Originally Posted by cheesehead

Can you quantify "few", "more", "significant", and so forth?

Another alternative to an automatic GCD check is a manual one, where you can click to check the GCD at any point you want.